Designing Empathetic, Resource-Aware AI Agents at Scale
summary
Argues for a concrete governance OS that wraps agents with a four-pillar governance model (policy, orchestration, feedback, override), describes an on-device deterministic enforcement prototype used at PhonePe that checks context against machine-readable user policies, blocks unauthorized spends, and surfaces a persistent cancel button to kill processes. Explains the need for an auditable feedback loop (human-readable proofs and logs) and reports the prototype neutralizes trust deficits enough to scale to over 100 million devices.
AI-written summary · about
chapters (8)
- 0:00The autonomy dilemma
- 2:43Redefining empathy
- 5:39The policy layer
- 10:25The orchestration layer
- 13:46Feedback and override layers
- 20:54Governance OS approach
- 24:19The automotive analogy
- 25:41Closing and Q&A
related talks
24mAI Governance for developers: A practical framework for deploying autonomous coding agents safely…
34mAccountability in the Age of AI Agents
13mKeynote: Agentic AI Is Running Your Infrastructure- 24mAI Governance for developers: A practical framework for deploying autonomous coding agents safely…
30mAgentic AI in Action: From Framework to Deployment
30mAI Agents Unchained
33mGoverning the Ungovernable: Security and Compliance for AI Agents in Open Source Projec...
57mDesigning Human-AI Experience